question on equipoise calculation

[ChungsooKim]

Hi, Currently, it seems to calculate the equipoise proportion in the population (target + comparator), which I think will not be valid when the numbers of patients are unbalanced.

According to the original paper on empirical equipoise, don't we are necessary to calculate the proportion of equipoise for each group (target/comparator) when determining equipoise according to preference score?

It would be useful if the proportion of patients with empirical equipoise for each group can be presented. If I misunderstood, please ignore this and forgive me.

[schuemie]

The equipoise is computed across both target and comparator. I must admit the Walker's paper isn't very clear on this. But just remember: a preference score is just a transformation of the propensity score, which only makes sense for a pair of treatments. Even though the paper sometimes shows equipoise proportions next to a single drug, this really means the proportion across target and comparator when comparing to that drug. Walker's paper specifically proposes to evaluate comparisons using the equipoise metric (of preference score between 0.3 and 0.7)

[ChungsooKim]

Thank you for your prompt reply @schuemie !! Maybe what I wrote misunderstood you. I'm not denying computing preference scores for target-comparator pairs.

However, I just wonder if it is necessary to be presented separately by target/comparison groups when presenting the proportion in equipoise (%).

Within the current cohortMethod, the proportion on equipoise is calculated as 0.3 <= preference score <= 0.7 for the study population (T + C).

but the original paper calculates the ratio of preference between 0.3 and 0.7 in each group (T and C respectively) and defines empirical equipoise when both groups are 0.5 or higher.

I was concerned about the ambiguous situation below rather than the general one. The proportion in equipoise for all patients is (193+154) / (320+369) * 100 = 50.3%. However, the percentages are 60 (193/320100) % and 41.7 (154/369100)%, respectively.

Once again, If I misunderstood, please forgive and correct me!

[schuemie]

Ah, I see you're right. I always misread this paper (probably because table 2 refers to 'comparisons', but then lists individual drugs per row). But indeed, it explicitly says:

Accept drug pairs as emerging from empirical equipoise if at least half of the dispensings of each of the drugs are to patients with a preference score of between 0.3 and 0.7

In practice, it shouldn't matter much. As you can see in table 2, the proportion in equipoise tend to be almost the same in both groups. But in specific cases, such as your example, it does make a difference.

So we'll need to change the equipoise in our diagnostic from being the overall percent in equipoise to the minimum percent in equipoise in the target and comparator.